GestaltMML: Enhancing Rare Genetic Disease Diagnosis through Multimodal Machine Learning Combining Facial Images and Clinical Texts.

Individuals with suspected rare genetic disorders often undergo multiple clinical evaluations, imaging studies, laboratory tests and genetic tests, to find a possible answer over a prolonged period of time. Addressing this "diagnostic odyssey" thus has substantial clinical, psychosocial, and economic benefits. Many rare genetic diseases have distinctive facial features, which can be used by artificial intelligence algorithms to facilitate clinical diagnosis, in prioritizing candidate diseases to be further examined by lab tests or genetic assays, or in helping the phenotype-driven reinterpretation of genome/exome sequencing data. Existing methods using frontal facial photos were built on conventional Convolutional Neural Networks (CNNs), rely exclusively on facial images, and cannot capture non-facial phenotypic traits and demographic information essential for guiding accurate diagnoses. Here we introduce GestaltMML, a multimodal machine learning (MML) approach solely based on the Transformer architecture. It integrates facial images, demographic information (age, sex, ethnicity), and clinical notes (optionally, a list of Human Phenotype Ontology terms) to improve prediction accuracy. Furthermore, we also evaluated GestaltMML on a diverse range of datasets, including 528 diseases from the GestaltMatcher Database, several in-house datasets of Beckwith-Wiedemann syndrome (BWS, over-growth syndrome with distinct facial features), Sotos syndrome (overgrowth syndrome with overlapping features with BWS), NAA10-related neurodevelopmental syndrome, Cornelia de Lange syndrome (multiple malformation syndrome), and KBG syndrome (multiple malformation syndrome). Our results suggest that GestaltMML effectively incorporates multiple modalities of data, greatly narrowing candidate genetic diagnoses of rare diseases and may facilitate the reinterpretation of genome/exome sequencing data.

Next-generation phenotyping in Nigerian children with Cornelia de Lange syndrome.

Next-generation phenotyping (NGP) can be used to compute the similarity of dysmorphic patients to known syndromic diseases. So far, the technology has been evaluated in variant prioritization and classification, providing evidence for pathogenicity if the phenotype matched with other patients with a confirmed molecular diagnosis. In a Nigerian cohort of individuals with facial dysmorphism, we used the NGP tool GestaltMatcher to screen portraits prior to genetic testing and subjected individuals with high similarity scores to exome sequencing (ES). Here, we report on two individuals with global developmental delay, pulmonary artery stenosis, and genital and limb malformations for whom GestaltMatcher yielded Cornelia de Lange syndrome (CdLS) as the top hit. ES revealed a known pathogenic nonsense variant, NM_133433.4: c.598C>T; p.(Gln200*), as well as a novel frameshift variant c.7948dup; p.(Ile2650Asnfs*11) in NIPBL. Our results suggest that NGP can be used as a screening tool and thresholds could be defined for achieving high diagnostic yields in ES. Training the artificial intelligence (AI) with additional cases of the same ethnicity might further increase the positive predictive value of GestaltMatcher.

Impact of Synbiotic Intake on Liver Metabolism in Metabolically Healthy Participants and Its Potential Preventive Effect on Metabolic-Dysfunction-Associated Fatty Liver Disease (MAFLD): A Randomized, Placebo-Controlled, Double-Blinded Clinical Trial.

Synbiotics modulate the gut microbiome and contribute to the prevention of liver diseases such as metabolic-dysfunction-associated fatty liver disease (MAFLD). This study aimed to evaluate the effect of a randomized, placebo-controlled, double-blinded seven-week intervention trial on the liver metabolism in 117 metabolically healthy male participants. Anthropometric data, blood parameters, and stool samples were analyzed using linear mixed models. After seven weeks of intervention, there was a significant reduction in alanine aminotransferase (ALT) in the synbiotic group compared to the placebo group (-14.92%, CI: -26.60--3.23%, p = 0.013). A stratified analysis according to body fat percentage revealed a significant decrease in ALT (-20.70%, CI: -40.88--0.53%, p = 0.045) in participants with an elevated body fat percentage. Further, a significant change in microbiome composition (1.16, CI: 0.06-2.25, p = 0.039) in this group was found, while the microbial composition remained stable upon intervention in the group with physiological body fat. The 7-week synbiotic intervention reduced ALT levels, especially in participants with an elevated body fat percentage, possibly due to modulation of the gut microbiome. Synbiotic intake may be helpful in delaying the progression of MAFLD and could be used in addition to the recommended lifestyle modification therapy.

Enhancing Variant Prioritization in VarFish through On-Premise Computational Facial Analysis.

Genomic variant prioritization is crucial for identifying disease-associated genetic variations. Integrating facial and clinical feature analyses into this process enhances performance. This study demonstrates the integration of facial analysis (GestaltMatcher) and Human Phenotype Ontology analysis (CADA) within VarFish, an open-source variant analysis framework. Challenges related to non-open-source components were addressed by providing an open-source version of GestaltMatcher, facilitating on-premise facial analysis to address data privacy concerns. Performance evaluation on 163 patients recruited from a German multi-center study of rare diseases showed PEDIA's superior accuracy in variant prioritization compared to individual scores. This study highlights the importance of further benchmarking and future integration of advanced facial analysis approaches aligned with ACMG guidelines to enhance variant classification.

Extending DeepTrio for sensitive detection of complex de novo mutation patterns.

De novo mutations (DNMs), and among them clustered DNMs within 20 bp of each other (cDNMs) are known to be a potential cause of genetic disorders. However, identifying DNM in whole genome sequencing (WGS) data is a process that often suffers from low specificity. We propose a deep learning framework for DNM and cDNM detection in WGS data based on Google's DeepTrio software for variant calling, which considers regions of 110 bp up- and downstream from possible variants to take information from the surrounding region into account. We trained a model each for the DNM and cDNM detection tasks and tested it on data generated on the HiSeq and NovaSeq platforms. In total, the model was trained on 82 WGS trios generated on the NovaSeq and 16 on the HiSeq. For the DNM detection task, our model achieves a sensitivity of 95.7% and a precision of 89.6%. The extended model adds confidence information for cDNMs, in addition to standard variant classes and DNMs. While this causes a slight drop in DNM sensitivity (91.96%) and precision (90.5%), on HG002 cDNMs can be isolated from other variant classes in all cases (5 out of 5) with a precision of 76.9%. Since the model emits confidence probabilities for each variant class, it is possible to fine-tune cutoff thresholds to allow users to select a desired trade-off between sensitivity and specificity. These results show that DeepTrio can be retrained to identify complex mutational signatures with only little modification effort.

Comparison of clinical geneticist and computer visual attention in assessing genetic conditions.

Artificial intelligence (AI) for facial diagnostics is increasingly used in the genetics clinic to evaluate patients with potential genetic conditions. Current approaches focus on one type of AI called Deep Learning (DL). While DL- based facial diagnostic platforms have a high accuracy rate for many conditions, less is understood about how this technology assesses and classifies (categorizes) images, and how this compares to humans. To compare human and computer attention, we performed eye-tracking analyses of geneticist clinicians (n = 22) and non-clinicians (n = 22) who viewed images of people with 10 different genetic conditions, as well as images of unaffected individuals. We calculated the Intersection-over-Union (IoU) and Kullback-Leibler divergence (KL) to compare the visual attentions of the two participant groups, and then the clinician group against the saliency maps of our deep learning classifier. We found that human visual attention differs greatly from DL model's saliency results. Averaging over all the test images, IoU and KL metric for the successful (accurate) clinician visual attentions versus the saliency maps were 0.15 and 11.15, respectively. Individuals also tend to have a specific pattern of image inspection, and clinicians demonstrate different visual attention patterns than non-clinicians (IoU and KL of clinicians versus non-clinicians were 0.47 and 2.73, respectively). This study shows that humans (at different levels of expertise) and a computer vision model examine images differently. Understanding these differences can improve the design and use of AI tools, and lead to more meaningful interactions between clinicians and AI technologies.

ZSCAN10 deficiency causes a neurodevelopmental disorder with characteristic oto-facial malformations.

Neurodevelopmental disorders are major indications for genetic referral and have been linked to more than 1,500 loci including genes encoding transcriptional regulators. The dysfunction of transcription factors often results in characteristic syndromic presentations, however, at least half of these patients lack a genetic diagnosis. The implementation of machine learning approaches has the potential to aid in the identification of new disease genes and delineate associated phenotypes. Next generation sequencing was performed in seven affected individuals with neurodevelopmental delay and dysmorphic features. Clinical characterization included reanalysis of available neuroimaging datasets and 2D portrait image analysis with GestaltMatcher. The functional consequences of ZSCAN10 loss were modelled in mouse embryonic stem cells (mESC), including a knock-out and a representative ZSCAN10 protein truncating variant. These models were characterized by gene expression and Western blot analyses, chromatin immunoprecipitation and quantitative PCR (ChIP-qPCR), and immunofluorescence staining. Zscan10 knockout mouse embryos were generated and phenotyped. We prioritized bi-allelic ZSCAN10 loss-of-function variants in seven affected individuals from five unrelated families as the underlying molecular cause. RNA-Seq analyses in Zscan10-/- mESCs indicated dysregulation of genes related to stem cell pluripotency. In addition, we established in mESCs the loss-of-function mechanism for a representative human ZSCAN10 protein truncating variant by showing alteration of its expression levels and subcellular localization, interfering with its binding to DNA enhancer targets. Deep phenotyping revealed global developmental delay, facial asymmetry, and malformations of the outer ear as consistent clinical features. Cerebral MRI showed dysplasia of the semicircular canals as an anatomical correlate of sensorineural hearing loss. Facial asymmetry was confirmed as a clinical feature by GestaltMatcher and was recapitulated in the Zscan10 mouse model along with inner and outer ear malformations. Our findings provide evidence of a novel syndromic neurodevelopmental disorder caused by bi-allelic loss-of-function variants in ZSCAN10.

Role of CAMK2D in neurodevelopment and associated conditions.

The calcium/calmodulin-dependent protein kinase type 2 (CAMK2) family consists of four different isozymes, encoded by four different genes-CAMK2A, CAMK2B, CAMK2G, and CAMK2D-of which the first three have been associated recently with neurodevelopmental disorders. CAMK2D is one of the major CAMK2 proteins expressed in the heart and has been associated with cardiac anomalies. Although this CAMK2 isoform is also known to be one of the major CAMK2 subtypes expressed during early brain development, it has never been linked with neurodevelopmental disorders until now. Here we show that CAMK2D plays an important role in neurodevelopment not only in mice but also in humans. We identified eight individuals harboring heterozygous variants in CAMK2D who display symptoms of intellectual disability, delayed speech, behavioral problems, and dilated cardiomyopathy. The majority of the variants tested lead to a gain of function (GoF), which appears to cause both neurological problems and dilated cardiomyopathy. In contrast, loss-of-function (LoF) variants appear to induce only neurological symptoms. Together, we describe a cohort of individuals with neurodevelopmental disorders and cardiac anomalies, harboring pathogenic variants in CAMK2D, confirming an important role for the CAMK2D isozyme in both heart and brain function.

Prevalence and determinants of subretinal drusenoid deposits in patients' first-degree relatives.

PURPOSE: Subretinal drusenoid deposits (SDDs) are distinct extracellular alteration anterior to the retinal pigment epithelium (RPE). Given their commonly uniform phenotype, a hereditary predisposition seems likely. Hence, we aim to investigate prevalence and determinants in patients' first-degree relatives. METHODS: We recruited SDD outpatients at their visits to our clinic and invited their relatives. We performed a full ophthalmic examination including spectral domain-optical coherence tomography (SD-OCT) and graded presence, disease stage of SDD as well as percentage of infrared (IR) en face area affected by SDD. Moreover, we performed genetic sequencing and calculated a polygenic risk score (PRS) for AMD. We conducted multivariable regression models to assess potential determinants of SDD and associations of SDD with PRS. RESULTS: We included 195 participants, 123 patients (mean age 81.4 ± 7.2 years) and 72 relatives (mean age 52.2 ± 14.2 years), of which 7 presented SDD, resulting in a prevalence of 9.7%. We found older age to be associated with SDD presence and area in the total cohort and a borderline association of higher body mass index (BMI) with SDD presence in the relatives. Individuals with SDD tended to have a higher PRS, which, however, was not statistically significant in the multivariable regression. CONCLUSION: Our study indicates a potential hereditary aspect of SDD and confirms the strong association with age. Based on our results, relatives of SDD patients ought to be closely monitored for retinal alterations, particularly at an older age. Further longitudinal studies with larger sample size and older relatives are needed to confirm or refute our findings.

Deeplasia: deep learning for bone age assessment validated on skeletal dysplasias.

BACKGROUND: Skeletal dysplasias collectively affect a large number of patients worldwide. Most of these disorders cause growth anomalies. Hence, evaluating skeletal maturity via the determination of bone age (BA) is a useful tool. Moreover, consecutive BA measurements are crucial for monitoring the growth of patients with such disorders, especially for timing hormonal treatment or orthopedic interventions. However, manual BA assessment is time-consuming and suffers from high intra- and inter-rater variability. This is further exacerbated by genetic disorders causing severe skeletal malformations. While numerous approaches to automate BA assessment have been proposed, few are validated for BA assessment on children with skeletal dysplasias. OBJECTIVE: We present Deeplasia, an open-source prior-free deep-learning approach designed for BA assessment specifically validated on patients with skeletal dysplasias. MATERIALS AND METHODS: We trained multiple convolutional neural network models under various conditions and selected three to build a precise model ensemble. We utilized the public BA dataset from the Radiological Society of North America (RSNA) consisting of training, validation, and test subsets containing 12,611, 1,425, and 200 hand and wrist radiographs, respectively. For testing the performance of our model ensemble on dysplastic hands, we retrospectively collected 568 radiographs from 189 patients with molecularly confirmed diagnoses of seven different genetic bone disorders including achondroplasia and hypochondroplasia. A subset of the dysplastic cohort (149 images) was used to estimate the test-retest precision of our model ensemble on longitudinal data. RESULTS: The mean absolute difference of Deeplasia for the RSNA test set (based on the average of six different reference ratings) and dysplastic set (based on the average of two different reference ratings) were 3.87 and 5.84 months, respectively. The test-retest precision of Deeplasia on longitudinal data (2.74 months) is estimated to be similar to a human expert. CONCLUSION: We demonstrated that Deeplasia is competent in assessing the age and monitoring the development of both normal and dysplastic bones.

Understanding recessive disease risk in multi-ethnic populations with different degrees of consanguinity.

Population medical genetics aims at translating clinically relevant findings from recent studies of large cohorts into healthcare for individuals. Genetic counseling concerning reproductive risks and options is still mainly based on family history, and consanguinity is viewed to increase the risk for recessive diseases regardless of the demographics. However, in an increasingly multi-ethnic society with diverse approaches to partner selection, healthcare professionals should also sharpen their intuition for the influence of different mating schemes in non-equilibrium dynamics. We, therefore, revisited the so-called out-of-Africa model and studied in forward simulations with discrete and not overlapping generations the effect of inbreeding on the average number of recessive lethals in the genome. We were able to reproduce in both frameworks the drop in the incidence of recessive disorders, which is a transient phenomenon during and after the growth phase of a population, and therefore showed their equivalence. With the simulation frameworks, we also provide the means to study and visualize the effect of different kin sizes and mating schemes on these parameters for educational purposes.

Delineation of the adult phenotype of Coffin-Siris syndrome in 35 individuals.

Coffin-Siris syndrome (CSS) is a rare multisystemic autosomal dominant disorder. Since 2012, alterations in genes of the SWI/SNF complex were identified as the molecular basis of CSS, studying largely pediatric cohorts. Therefore, there is a lack of information on the phenotype in adulthood, particularly on the clinical outcome in adulthood and associated risks. In an international collaborative effort, data from 35 individuals ≥ 18 years with a molecularly ascertained CSS diagnosis (variants in ARID1B, ARID2, SMARCA4, SMARCB1, SMARCC2, SMARCE1, SOX11, BICRA) using a comprehensive questionnaire was collected. Our results indicate that overweight and obesity are frequent in adults with CSS. Visual impairment, scoliosis, and behavioral anomalies are more prevalent than in published pediatric or mixed cohorts. Cognitive outcomes range from profound intellectual disability (ID) to low normal IQ, with most individuals having moderate ID. The present study describes the first exclusively adult cohort of CSS individuals. We were able to delineate some features of CSS that develop over time and have therefore been underrepresented in previously reported largely pediatric cohorts, and provide recommendations for follow-up.

Trans-ancestry polygenic models for the prediction of LDL blood levels: an analysis of the United Kingdom Biobank and Taiwan Biobank.

Polygenic risk score (PRS) predictions often show bias toward the population of available genome-wide association studies (GWASs), which is typically of European ancestry. This study aimed to assess the performance differences of ancestry-specific PRS and test the implementation of multi-ancestry PRS to enhance the generalizability of low-density lipoprotein (LDL) cholesterol predictions in the East Asian (EAS) population. In this study, we computed ancestry-specific and multi-ancestry PRSs for LDL using data obtained from the Global Lipid Genetics Consortium, while accounting for population-specific linkage disequilibrium patterns using the PRS-CSx method in the United Kingdom Biobank dataset (UKB, n = 423,596) and Taiwan Biobank dataset (TWB, n = 68,978). Population-specific PRSs were able to predict LDL levels better within the target population, whereas multi-ancestry PRSs were more generalizable. In the TWB dataset, covariate-adjusted R 2 values were 9.3% for ancestry-specific PRS, 6.7% for multi-ancestry PRS, and 4.5% for European-specific PRS. Similar trends (8.6%, 7.8%, and 6.2%) were observed in the smaller EAS population of the UKB (n = 1,480). Consistent with R 2 values, PRS stratification in EAS regions (TWB) effectively captured a heterogenous variability in LDL blood cholesterol levels across PRS strata. The mean difference in LDL levels between the lowest and highest EAS-specific PRS (EAS_PRS) deciles was 0.82, compared to 0.59 for European-specific PRS (EUR_PRS) and 0.76 for multi-ancestry PRS. Notably, the mean LDL values in the top decile of multi-ancestry PRS were comparable to those of EAS_PRS (3.543 vs. 3.541, p = 0.86). Our analysis of the PRS prediction model for LDL cholesterol further supports the issue of PRS generalizability across populations. Our targeted analysis of the EAS population revealed that integrating non-European genotyping data with a powerful European-based GWAS can enhance the generalizability of LDL PRS.

Facilitating the Molecular Diagnosis of Rare Genetic Disorders Through Facial Phenotypic Scores.

With recent advances in computer vision, many applications based on artificial intelligence have been developed to facilitate the diagnosis of rare genetic disorders through the analysis of patients' two-dimensional frontal images. Some of these have been implemented on online platforms with user-friendly interfaces and provide facial analysis services, such as Face2Gene. However, users cannot run the facial analysis processes in house because the training data and the trained models are unavailable. This article therefore provides an introduction, designed for users with programming backgrounds, to the use of the open-source GestaltMatcher approach to run facial analysis in their local environment. The Basic Protocol provides detailed instructions for applying for access to the trained models and then performing facial analysis to obtain a prediction score for each of the 595 genes in the GestaltMatcher Database. The prediction results can then be used to narrow down the search space of disease-causing mutations or further connect with a variant-prioritization pipeline. © 2023 The Authors. Current Protocols published by Wiley Periodicals LLC. Basic Protocol: Using the open-source GestaltMatcher approach to perform facial analysis.

Analysis of 72,469 UK Biobank exomes links rare variants to male-pattern hair loss.

Male-pattern hair loss (MPHL) is common and highly heritable. While genome-wide association studies (GWAS) have generated insights into the contribution of common variants to MPHL etiology, the relevance of rare variants remains unclear. To determine the contribution of rare variants to MPHL etiology, we perform gene-based and single-variant analyses in exome-sequencing data from 72,469 male UK Biobank participants. While our population-level risk prediction suggests that rare variants make only a minor contribution to general MPHL risk, our rare variant collapsing tests identified a total of five significant gene associations. These findings provide additional evidence for previously implicated genes (EDA2R, WNT10A) and highlight novel risk genes at and beyond GWAS loci (HEPH, CEPT1, EIF3F). Furthermore, MPHL-associated genes are enriched for genes considered causal for monogenic trichoses. Together, our findings broaden the MPHL-associated allelic spectrum and provide insights into MPHL pathobiology and a shared basis with monogenic hair loss disorders.

Gene-based burden scores identify rare variant associations for 28 blood biomarkers.

BACKGROUND: A relevant part of the genetic architecture of complex traits is still unknown; despite the discovery of many disease-associated common variants. Polygenic risk score (PRS) models are based on the evaluation of the additive effects attributable to common variants and have been successfully implemented to assess the genetic susceptibility for many phenotypes. In contrast, burden tests are often used to identify an enrichment of rare deleterious variants in specific genes. Both kinds of genetic contributions are typically analyzed independently. Many studies suggest that complex phenotypes are influenced by both low effect common variants and high effect rare deleterious variants. The aim of this paper is to integrate the effect of both common and rare functional variants for a more comprehensive genetic risk modeling. METHODS: We developed a framework combining gene-based scores based on the enrichment of rare functionally relevant variants with genome-wide PRS based on common variants for association analysis and prediction models. We applied our framework on UK Biobank dataset with genotyping and exome data and considered 28 blood biomarkers levels as target phenotypes. For each biomarker, an association analysis was performed on full cohort using gene-based scores (GBS). The cohort was then split into 3 subsets for PRS construction and feature selection, predictive model training, and independent evaluation, respectively. Prediction models were generated including either PRS, GBS or both (combined). RESULTS: Association analyses of the cohort were able to detect significant genes that were previously known to be associated with different biomarkers. Interestingly, the analyses also revealed heterogeneous effect sizes and directionality highlighting the complexity of the blood biomarkers regulation. However, the combined models for many biomarkers show little or no improvement in prediction accuracy compared to the PRS models. CONCLUSION: This study shows that rare variants play an important role in the genetic architecture of complex multifactorial traits such as blood biomarkers. However, while rare deleterious variants play a strong role at an individual level, our results indicate that classical common variant based PRS might be more informative to predict the genetic susceptibility at the population level.

Integration of EpiSign, facial phenotyping, and likelihood ratio interpretation of clinical abnormalities in the re-classification of an ARID1B missense variant.

Heterozygous ARID1B variants result in Coffin-Siris syndrome. Features may include hypoplastic nails, slow growth, characteristic facial features, hypotonia, hypertrichosis, and sparse scalp hair. Most reported cases are due to ARID1B loss of function variants. We report a boy with developmental delay, feeding difficulties, aspiration, recurrent respiratory infections, slow growth, and hypotonia without a clinical diagnosis, where a previously unreported ARID1B missense variant was classified as a variant of uncertain significance. The pathogenicity of this variant was refined through combined methodologies including genome-wide methylation signature analysis (EpiSign), Machine Learning (ML) facial phenotyping, and LIRICAL. Trio exome sequencing and EpiSign were performed. ML facial phenotyping compared facial images using FaceMatch and GestaltMatcher to syndrome-specific libraries to prioritize the trio exome bioinformatic pipeline gene list output. Phenotype-driven variant prioritization was performed with LIRICAL. A de novo heterozygous missense variant, ARID1B p.(Tyr1268His), was reported as a variant of uncertain significance. The ACMG classification was refined to likely pathogenic by a supportive methylation signature, ML facial phenotyping, and prioritization through LIRICAL. The ARID1B genotype-phenotype has been expanded through an extended analysis of missense variation through genome-wide methylation signatures, ML facial phenotyping, and likelihood-ratio gene prioritization.

Human and computer attention in assessing genetic conditions.

Deep learning (DL) and other types of artificial intelligence (AI) are increasingly used in many biomedical areas, including genetics. One frequent use in medical genetics involves evaluating images of people with potential genetic conditions to help with diagnosis. A central question involves better understanding how AI classifiers assess images compared to humans. To explore this, we performed eye-tracking analyses of geneticist clinicians and non-clinicians. We compared results to DL-based saliency maps. We found that human visual attention when assessing images differs greatly from the parts of images weighted by the DL model. Further, individuals tend to have a specific pattern of image inspection, and clinicians demonstrate different visual attention patterns than non-clinicians.